Since wsl2 supports cuda, my gaming computer can run open source deep learning models so easily it’s stupid. I’m mainly using it to rip music from youtube and split it into stems for music production using Facebook demucs. I tinkered a bit with stable diffusion models a while back too. It’s pretty sweet, especially since windows sees the linux drive as just another directory, so my DAW can just bookmark it. It’s so seamless.
Win 11 is still garbage for privacy and ownership reasons though. MS can fuck a duck, but they make some pretty baller software.
I’ll emphasize the point that this goes for any kind of machine learning model that can benefit from CUDA, which means a large amount of gaming computers already meet the prerequisites for this. Installation is trivial (but requires some knowledge), and I hope to see more ML applications for hobbyists in the near future. Image generation and locally hosted GPT models come to mind.
Since wsl2 supports cuda, my gaming computer can run open source deep learning models so easily it’s stupid. I’m mainly using it to rip music from youtube and split it into stems for music production using Facebook demucs. I tinkered a bit with stable diffusion models a while back too. It’s pretty sweet, especially since windows sees the linux drive as just another directory, so my DAW can just bookmark it. It’s so seamless.
Win 11 is still garbage for privacy and ownership reasons though. MS can fuck a duck, but they make some pretty baller software.
That’s an interesting use case.
I’ll emphasize the point that this goes for any kind of machine learning model that can benefit from CUDA, which means a large amount of gaming computers already meet the prerequisites for this. Installation is trivial (but requires some knowledge), and I hope to see more ML applications for hobbyists in the near future. Image generation and locally hosted GPT models come to mind.